Human Visual Perception Based Image Quality Assessment for Video Prediction

Abstract

Prediction is an important ability of robot to imitate human intelligence. In recent years, video prediction in the computer vision field provides the prediction-coding implementation for robots, which allows robots to predict a reasonable future scene based on a little continuous video information by generating multi-frame images. Since the images of the predicted generation vary based on different algorithms, it tends to cause different types and degrees of distortion, thereby reducing the prediction-oriented operability of robots. The image quality assessment(IQA) for video prediction task must consider whether it is consistent with human visual perception to judge the rationality, not just get some shallow information like PSNR or SSIM method. Based on the two alternative forced choice (2AFC) experiment of human visual perception, this paper proposes a perceptual assessment metric based on convolutional feature extraction networks for video prediction task. The assessment results are consistent with the law that the video prediction quality declines over time, and more correspond to the human perception. In addition, the VggNet used as feature extraction network is more sensitive to image quality, so it will be easier to assess similar images.